search

Autodesk / Aurora

Real-time GPU path tracing with an OpenUSD Hydra render delegate
Apache License 2.0
506 stars 31 forks source link

Specify thread group dimensions for compute shaders #30

Closed pixnblox closed 1 year ago

pixnblox commented 1 year ago

This modifies the DirectX compute shaders (accumulation and post-processing) to specify thread group dimensions. This is done to improve GPU occupancy, and thus performance. Specifically, this reduces the time required to run those compute shaders, which improves overall rendering performance. For simple scenarios, this can actually double performance as measured in Plasma. In more typical scenarios, this will increase performance by around 10%.

This change was suggested by Neil Bickford, a developer technology engineer from NVIDIA.

gareth-morgan-autodesk commented 1 year ago

Please use "squash and merge" (this should be the default/only option, if I've setup the settings correctly)